Modelling the Imports and Exports Data of Türkiye Employing Autoregressive Artificial Neural Networks

Modelling imports and exports data is crucial for understanding and forecasting economic trends, as it helps policymakers and businesses make informed decisions on trade policies and strategies. Accurate models can reveal underlying patterns in trade flows, enabling better management of economic stability and growth. This study presents the modelling of Türkiye's total imports and exports using artificial neural networks (ANNs). The data, obtained from official sources, exhibited significant nonlinearity and non-stationarity, prompting the development of an ANN with three hidden layers, each containing ten neurons. The autoregressive ANN model utilized lagged values of imports and exports as inputs, and was trained separately for both datasets. The training process converged after 686 epochs for exports and 522 epochs for imports, demonstrating the model's effectiveness. The comparison of actual data with ANN predictions revealed a high level of accuracy, further validated by performance metrics such as the coefficient of determination, mean absolute percentage error, mean absolute error, and root mean square error. With coefficients of determination exceeding 0.85, the model's accuracy and robustness were confirmed. Notably, the same ANN architecture was employed for both imports and exports, highlighting the model's versatility. The study suggests that the developed autoregressive ANN model is applicable to other time series data characterized by strong nonlinearity.